For years, the typical analytics workflow has looked something like this: A marketer or analyst opens a dashboard, notices something unusual, investigates the data, develops a hypothesis, decides what to do, takes action, and then measures the results.
At least, that’s how it’s supposed to work.
In reality, most analysts are already overloaded. They spend much of their time collecting data, maintaining tracking, building reports, responding to requests, and explaining what happened. There often isn’t enough time to proactively monitor every metric, investigate every anomaly, and determine which issues actually deserve attention.
As a result, important changes can sit unnoticed in dashboards until someone happens to look.
This is where I think AI can fundamentally change the analytics workflow.
Instead of waiting for an analyst to discover a problem, AI agents can continuously monitor the data, identify unusual changes, investigate possible causes, prioritize what matters, and bring the important issues to us.
The analyst doesn’t disappear. The analyst moves from constantly looking for problems to evaluating, investigating, and acting on the problems that actually matter.
Eventually, many of these agents won’t stop at recommendations, they will take action themselves. That sounds incredibly powerful but it also creates a problem that I think deserves much more attention:
What happens when the data the AI is using is wrong?
AI Can Reason Correctly and Still Make the Wrong Decision
Imagine an AI marketing agent monitoring an ecommerce business see that yesterday revenue dropped by 30%.
The agent looks at the data and sees:
- Traffic is roughly unchanged.
- Advertising spend is unchanged.
- Reported purchases are down significantly.
- Reported revenue is down significantly.
Based on this information, the agent concludes that marketing performance has deteriorated. It recommends reducing advertising spend. That sounds reasonable, except there is one problem.
Revenue didn’t actually fall 30%.
Someone published a change in Google Tag Manager yesterday and accidentally broke the purchase tracking.
The AI may have reasoned perfectly but the evidence was wrong.
This is an important distinction.
We spend a lot of time talking about AI hallucinations and whether an AI model will generate an incorrect answer.
But there is another problem that may become equally important in marketing: Correct reasoning based on incorrect measurement.
And as AI moves from answering questions to taking actions, the consequences of bad measurement become much larger.
The Analytics Workflow Is Changing
Historically, analytics has largely been something we query.
We open GA4, Adobe Analytics, Amplitude, Mixpanel, a BI tool, or another analytics platform and ask questions about what happened.
AI is beginning to reverse that relationship. Instead of us constantly going to analytics, analytics can increasingly come to us. The emerging workflow looks more like this:
Observe → Understand → Decide → Act → Measure → Learn
Each stage answers a different question.
Observe: What happened?
The system continuously monitors marketing, customer, product, and business data.
- Did revenue decline?
- Did conversion rate change?
- Did organic traffic suddenly increase?
- Did checkout abandonment spike?
- Did a campaign stop generating qualified leads?
This sounds straightforward, but there is a critical question underneath all of it:
Can we trust what the system is observing?
If purchase tracking is broken, consent settings changed, UTMs disappeared, events are duplicated, or attribution logic changed, the observation itself may be wrong.
Everything that follows can then be wrong as well.
Understand: What Actually Matters?
A typical organization can generate hundreds or thousands of signals every day.
- Traffic changed.
- Revenue changed.
- A campaign changed.
- A conversion rate changed.
- A new GTM version was published.
- An event stopped firing.
- An audience changed.
- A landing page was updated.
Humans already struggle with this volume of information.
AI has the potential to help determine which changes actually deserve attention.
Instead of sending ten alerts, an intelligent system might say:
Here are the three things you should investigate today.
But identifying an anomaly is only part of the job.
The harder question is:
Why did it happen?
A revenue decline might be caused by poor campaign performance, or a checkout problem, or seasonality, or inventory, or a website change, or broken analytics, etc.
This is where AI becomes much more interesting than simply generating summaries of dashboards.
Decide: AI Should Reason Over Trusted Facts
Once the system understands what happened, it has to decide what should be done. This is where I believe we need an important separation between measurement and reasoning.
Consider seemingly simple questions such as:
- What counts as a conversion?
- What is revenue?
- What qualifies as a lead?
- How is a marketing channel defined?
- Which attribution model should be used?
These definitions should not be invented by an LLM every time someone asks a question. They should come from a trusted measurement layer. The analytics system should establish the facts.
AI should reason over those facts.
For example, a measurement system might establish:
- Revenue is 24% below the expected range.
- Website sessions are normal.
- Purchase events are down 27%.
- Average order value is unchanged.
- A GTM version was published yesterday.
- The purchase-event audit is showing a warning.
Now AI has something useful to reason about.
Instead of calculating the underlying truth itself, it can ask:
- What is the most likely explanation?
- What should the marketer investigate first?
- How urgent is the issue?
- What evidence supports the recommendation?
That is a much better role for AI.
Act: Bad Data Becomes More Dangerous When AI Can Take Action
Today, bad analytics might result in someone making a bad decision after looking at a dashboard.
Tomorrow, an AI agent might make that decision automatically.
Imagine an agent that can:
- Change Google Ads budgets.
- Pause campaigns.
- Modify audiences.
- Personalize website experiences.
- Launch email campaigns.
- Change offers.
- Adjust bidding strategies.
Now return to our broken purchase-tag example.
The agent sees reported revenue collapse and automatically cuts advertising spend.
The tracking problem has now become a business problem.
This is why I believe measurement integrity becomes more important in an AI-driven marketing environment, not less important.
The faster machines can make decisions, the more important it becomes to make sure the evidence behind those decisions is trustworthy.
Measure: Did the AI’s Decision Actually Work?
There is another challenge.
Suppose an AI agent recommends changing a campaign. The change is made. Revenue increases 12% the following week. Was the AI right? Maybe.
But revenue could have increased because of seasonality, a promotion, changes in demand, another marketing channel, or dozens of other factors.
Simply observing that a metric improved after an AI recommendation does not prove that the recommendation caused the improvement.
This is where experimentation and incrementality become increasingly important.
We need to ask:
Did the action actually cause the outcome?
That may require controlled experiments, holdout groups, incrementality testing, causal analysis, or other measurement approaches.
As marketing becomes more automated, measurement can’t stop at reporting what happened. It needs to help determine what actually worked.
Learn: What Should the System Remember?
The final part of the loop may ultimately be one of the most valuable.
Imagine the purchase-tag problem happens again six months later. Should the system start its investigation from scratch? Ideally, no.
It should remember:
A similar revenue decline occurred previously. Traffic remained stable. A GTM change had recently occurred. Purchase tracking failed. The issue was fixed by correcting the tag.
Now the system has organizational memory.
The next time similar signals appear, it might say:
Revenue appears to be down 28%, but before changing marketing spend, verify the purchase implementation. A similar pattern previously resulted from a GTM tracking issue.
That’s very different from a traditional dashboard.
The system isn’t simply reporting data, it’s learning how the business behaves.
Where Customer Data Fits
Analytics data alone won’t be enough for many AI marketing decisions.
An agent may also need to understand:
- Who the customer is.
- What products they have purchased.
- Which campaigns they have interacted with.
- Whether they have consented to certain uses of their data.
- Which audience they belong to.
- Their relationship with the company.
- Business rules governing how they should be treated.
This is where CRM systems, CDPs, data warehouses, and other customer-data platforms could become increasingly important.
Their role may evolve beyond simply creating customer profiles or audiences.
They could become part of the trusted context layer that AI agents use to make decisions.
You can think of the emerging architecture roughly like this:
Analytics → trusted behavioral signals
CRM/CDP → trusted customer context
Business systems → objectives, constraints, and outcomes
AI → reasoning and decision-making
Marketing platforms → execution
Measurement → determine whether it worked
And then the loop starts again.
What We’re Exploring With GA Auditor
This is also one of the questions we’re beginning to explore with GA Auditor at Optizent.
GA Auditor already examines GA4 and Google Tag Manager implementations and monitors analytics data for potential problems.
Today, systems like this primarily identify issues.
For example:
Revenue dropped significantly.
Engagement changed unexpectedly.
A tracking implementation has a problem.
A GTM change occurred.
The next question we’re exploring is more interesting:
Can AI take these trusted signals and determine what someone should investigate first?
Instead of receiving numerous alerts, imagine receiving something like:
Three Things You Should Investigate Today
1. Verify purchase tracking — High priority
Reported revenue fell 27%, but traffic and average order value remain within their normal ranges. A GTM change was published yesterday and the purchase-event audit is showing a warning.
Recommended action: Verify the purchase event before changing marketing spend.
That’s much closer to a decision-support system than an analytics alert.
And it creates an interesting experiment for us:
Can an AI reliably distinguish a marketing problem from a measurement problem?
We can deliberately create scenarios involving both.
Some might represent genuine business problems:
- Conversion rate deterioration.
- Poor campaign performance.
- Checkout friction.
- Audience fatigue.
- Landing-page problems.
Others might represent measurement problems:
- Broken purchase tracking.
- Duplicate events.
- Consent configuration changes.
- Incorrect UTMs.
- GTM publishing mistakes.
- Attribution configuration changes.
Then we can evaluate whether the AI identifies the right type of problem and recommends the appropriate next action.
That’s the kind of practical AI experiment I think marketers and analysts should be running.
Trustworthy AI Starts With Trustworthy Measurement
There is enormous excitement around AI agents in marketing, and much of it is justified.
AI will likely become very good at helping us:
Understand. Decide. Act.
And eventually:
Learn.
But the complete system still looks like this:
Observe → Understand → Decide → Act → Measure → Learn
The quality of every stage depends on the stages around it.
If Observe is based on unreliable data, AI can make an intelligent decision based on a false premise.
If Measure can’t determine whether the action actually worked, the system can learn the wrong lesson.
That’s why I don’t think AI reduces the importance of analytics and measurement.
It increases it.
The future of marketing won’t simply be about building smarter AI agents. It will be about building AI systems that can make trustworthy decisions based on trustworthy evidence—and then reliably determine whether those decisions worked.
Start With Data You Can Trust
Before AI can make better marketing decisions, it needs reliable data to work with.
That starts with making sure your analytics implementation is working correctly and continues with monitoring it, because tracking can break at any time.
GA Auditor automatically monitors your GA4 and Google Tag Manager setup and helps identify tracking and data-quality issues before they turn into bad business decisions.
Set up automatic monitoring with GA Auditor
Need help improving your analytics, measurement strategy, or preparing your data for AI-driven decision-making? Contact Optizent